Improving the utility of locally differentially private protocols for longitudinal and multidimensional frequency estimates

Arcolezi, Héber H.; Couchot, Jean-François; Bouna, Béchara Al; Xiao, Xiaokui

doi:10.1016/j.dcan.2022.07.003

Cited by 9 publications

(12 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The three following subsections briefly present state-of-the-art longitudinal LDP mechanisms [3,6,16], which are based on GRR, LH, and UE mechanisms, i.e., same encoding, perturbation, and aggregation methods. Therefore, we focus on presenting the composition of two LDP mechanisms (i.e., M 1 •M 2 ) and both p * and q * parameters that are based on the privacy guarantees ϵ ∞ , ϵ 1 .…”

Section: Ldp Mechanisms For Multiple Data Collectionsmentioning

confidence: 99%

“…Algorithm 2 Memoization-based procedure for longitudinal distribution estimation under LDP guarantees [3,6,16].…”

Section: Ldp Mechanisms For Multiple Data Collectionsmentioning

confidence: 99%

“…To address this issue, one line of research has largely explored the issue of improving the utility of LDP distribution estimation mechanisms [7,16,18,24,25,30], in which the data collector estimates the number of users for each possible value based on the sanitized data of the users. Distribution estimation is a primary objective of LDP [29], as it is a building block for more complex tasks (e.g., heavy hitter estimation [7], marginal estimation [9,17,20,23], and distribution monitoring [3,6,16]).…”

Section: Introductionmentioning

confidence: 99%

“…More precisely, we have experimentally assessed seven state-of-the-art LDP distribution estimation mechanisms, namely, Generalized Randomized Response (GRR) [18], Binary Local Hashing (BLH) [7], Optimal Local Hashing (OLH) [25], Symmetric Unary Encoding (SUE) [16], Optimal Unary Encoding (OUE) [25], Subset Selection (SS) [24,30], and Thresholding with Histogram Encoding (THE) [25]. In addition, we also extended the analysis to seven state-of-the-art memoizationbased LDP mechanisms, namely, Longitudinal GRR (L-GRR) [3], four Longitudinal Unary Encoding (L-UE) [3,16] mechanisms, and two Longitudinal Local Hashing (L-LH) [6] mechanisms. To further broaden the scope of our study, we have evaluated two popular utility metrics by varying the number of users n, the privacy guarantee ϵ, and the domain size k, with both synthetic and realworld datasets.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

On the Utility Gain of Iterative Bayesian Update for Locally Differentially Private Mechanisms

Arcolezi

Cerna

Palamidessi

2023

Lecture Notes in Computer Science

Self Cite

View full text Add to dashboard Cite

This paper investigates the utility gain of using Iterative Bayesian Update (IBU) for private discrete distribution estimation using data obfuscated with Locally Differentially Private (LDP) mechanisms. We compare the performance of IBU to Matrix Inversion (MI), a standard estimation technique, for seven LDP mechanisms designed for onetime data collection and for other seven LDP mechanisms designed for multiple data collections (e.g., RAPPOR). To broaden the scope of our study, we also varied the utility metric, the number of users n, the domain size k, and the privacy parameter ϵ, using both synthetic and real-world data. Our results suggest that IBU can be a useful post-processing tool for improving the utility of LDP mechanisms in different scenarios without any additional privacy cost. For instance, our experiments show that IBU can provide better utility than MI, especially in high privacy regimes (i.e., when ϵ is small). Our paper provides insights for practitioners to use IBU in conjunction with existing LDP mechanisms for more accurate and privacy-preserving data analysis. Finally, we implemented IBU for all fourteen LDP mechanisms into the state-of-the-art multi-freq-ldpy Python package (https://pypi.org/project/multi-freq-ldpy/) and open-sourced all our code used for the experiments as tutorials.

show abstract

Section: Ldp Mechanisms For Multiple Data Collectionsmentioning

confidence: 99%

“…Algorithm 2 Memoization-based procedure for longitudinal distribution estimation under LDP guarantees [3,6,16].…”

Section: Ldp Mechanisms For Multiple Data Collectionsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

On the Utility Gain of Iterative Bayesian Update for Locally Differentially Private Mechanisms

Arcolezi

Cerna

Palamidessi

2023

Lecture Notes in Computer Science

Self Cite

View full text Add to dashboard Cite

show abstract

“…Afrose et al (2021) [31] proposed local differential privacy for stream data using RAPPOR technique to obtain frequency estimation. Arcolezi et al (2021) [32] conducted a study on temporal data of multidimensional attribute using three state-of-the-art protocols. Wang et al (2021) [33] proposed a method to determine the optimal threshold with exponential mechanism to satisfy the data stream publishing problems.…”

Section: Plos Onementioning

confidence: 99%

Local differential privacy protection for wearable device data

Wang

et al. 2022

PLoS ONE

View full text Add to dashboard Cite

Personal data collected by wearable devices contains rich privacy. It is important to realize the personal privacy protection for user data without affecting the data collection of wearable device services. In order to protect users’ personal privacy, a collection scheme based on local differential privacy is proposed for the collected single attribute numerical stream data. At first, the stream data points collected by the wearable device are censored to identify the salient points, and the adaptive Laplacian mechanism is used to add noise to these salient points according to the assigned privacy budget; then the collector reconstructs and fits the stream data curve to the noise-added salient points, so as to protect the personal privacy of the data. This scheme is experimented on the heart rate dataset, and the results show that when the privacy budget is 0.5 (i.e., at higher privacy protection strength), the mean relative error is 0.12, which is 57.78% lower than the scheme of Kim et al. With the satisfaction of user privacy protection, the usability of mean value estimation of wearable device stream data is improved.

show abstract

Multi-Freq-LDPy: Multiple Frequency Estimation Under Local Differential Privacy in Python

Arcolezi

Couchot

Gambs

et al. 2022

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Improving the utility of locally differentially private protocols for longitudinal and multidimensional frequency estimates

Cited by 9 publications

References 13 publications

On the Utility Gain of Iterative Bayesian Update for Locally Differentially Private Mechanisms

On the Utility Gain of Iterative Bayesian Update for Locally Differentially Private Mechanisms

Local differential privacy protection for wearable device data

Multi-Freq-LDPy: Multiple Frequency Estimation Under Local Differential Privacy in Python

Contact Info

Product

Resources

About